RoquiDavid/Multimodal-Approach-to-Heritage-Preservation-in-the-Context-of-Climate-Change: CAA2025 - Code and Data

Code and experimental notebooks for the paper "A Multimodal Approach to Heritage Preservation in the Context of Climate Change", accepted at CAA 2025 (Computer Applications and Quantitative Methods in Archaeology). This release includes the full implementation of our lightweight multimodal architecture combining PerceiverIO with Adaptive Barlow Twins loss for fusing environmental sensor data and visual imagery to predict degradation severity at heritage sites. Experiments were conducted on data from Strasbourg Cathedral. Main results: 69.2% accuracy, outperforming standard multimodal baselines (VisualBERT, Transformer) by up to 50%.